PD37-03 MULTI-INSTITUTIONAL MACHINE LEARNING TOOL FOR PREDICTING HIGH RISK LESIONS ON 3 TESLA MULTIPARAMETRIC PROSTATE MRI

Abstract

You have accessJournal of UrologyImaging/Radiology: Uroradiology III1 Apr 2018PD37-03 MULTI-INSTITUTIONAL MACHINE LEARNING TOOL FOR PREDICTING HIGH RISK LESIONS ON 3 TESLA MULTIPARAMETRIC PROSTATE MRI Matthew Truong, Janet Kukreja, Soroush Rais-Bahrami, Nimrod Barashi, Bokai Wang, Zachary Nuffer, Ji Hae Park, Khoa Lam, Thomas Frye, Jeffrey Nix, John Thomas, Changyong Feng, Brian Chapin, John Davis, Gary Hollenberg, Aytekin Oto, Scott Eggener, Jean Joseph, Eric Weinberg, and Edward Messing Matthew TruongMatthew Truong More articles by this author , Janet KukrejaJanet Kukreja More articles by this author , Soroush Rais-BahramiSoroush Rais-Bahrami More articles by this author , Nimrod BarashiNimrod Barashi More articles by this author , Bokai WangBokai Wang More articles by this author , Zachary NufferZachary Nuffer More articles by this author , Ji Hae ParkJi Hae Park More articles by this author , Khoa LamKhoa Lam More articles by this author , Thomas FryeThomas Frye More articles by this author , Jeffrey NixJeffrey Nix More articles by this author , John ThomasJohn Thomas More articles by this author , Changyong FengChangyong Feng More articles by this author , Brian ChapinBrian Chapin More articles by this author , John DavisJohn Davis More articles by this author , Gary HollenbergGary Hollenberg More articles by this author , Aytekin OtoAytekin Oto More articles by this author , Scott EggenerScott Eggener More articles by this author , Jean JosephJean Joseph More articles by this author , Eric WeinbergEric Weinberg More articles by this author , and Edward MessingEdward Messing More articles by this author View All Author Informationhttps://doi.org/10.1016/j.juro.2018.02.1738AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookTwitterLinked InEmail INTRODUCTION AND OBJECTIVES There are no existing tools to determine the pre-test probability of detecting high risk prostate multiparametric MRI (mpMRI) lesions. Our objective was to develop and validate a machine learning tool to aid with patient selection prior to prostate mpMRI. METHODS Four tertiary care centers with mpMRI expertise were included (BiRCH Study Collaborative: Birmingham, Rochester, Chicago, and Houston). Prediction models were developed using 1269 patients who met mpMRI inclusion criteria: mpMRI performed in a biopsy naive patient; mpMRI performed after at least one prior negative 12-core transrectal ultrasound-guided biopsy; mpMRI performed during active surveillance for low risk PCa. Using age, PSA, and prostate volume, a support vector machine (SVM) model was developed for predicting the probability of harboring Prostate Imaging - Reporting and Data System (PIRADS) 4 or 5 lesions. Receiver operating characteristic (ROC) curves, calibration curves, and decision curves were generated to assess performance. Bias correction was performed using 10-fold cross validation. Prospective external validation was performed in 88 consecutive biopsy naive and prior negative biopsy patients undergoing 3 Tesla mpMRI. RESULTS For biopsy naive and prior negative biopsy patients (n = 811), the machine learning tool discriminated with an area under the curve (AUC) of 0.730 on internal validation. Moreover, excellent calibration and high net clinical benefit were observed across all risk thresholds. BiRCH was not clinically useful in active surveillance patients (n = 458) on decision curve analysis. The final model for use in biopsy naive and prior negative biopsy patients was developed on the Microsoft Azure Machine Learning platform and can be accessed at birch.azurewebsites.net. On prospective external validation (n = 88), the model discriminated with an AUC of 0.740 (95% confidence interval 0.624-0.856). CONCLUSIONS We developed and prospectively validated BiRCH, a machine learning tool that inputs readily available clinical variables (age, PSA, and prostate volume) and predicts accurately whether a patient will have high risk lesions. Using this online tool, clinicians can determine which patients will benefit most from prostate mpMRI. © 2018FiguresReferencesRelatedDetails Volume 199Issue 4SApril 2018Page: e731-e732 Advertisement Copyright & Permissions© 2018MetricsAuthor Information Matthew Truong More articles by this author Janet Kukreja More articles by this author Soroush Rais-Bahrami More articles by this author Nimrod Barashi More articles by this author Bokai Wang More articles by this author Zachary Nuffer More articles by this author Ji Hae Park More articles by this author Khoa Lam More articles by this author Thomas Frye More articles by this author Jeffrey Nix More articles by this author John Thomas More articles by this author Changyong Feng More articles by this author Brian Chapin More articles by this author John Davis More articles by this author Gary Hollenberg More articles by this author Aytekin Oto More articles by this author Scott Eggener More articles by this author Jean Joseph More articles by this author Eric Weinberg More articles by this author Edward Messing More articles by this author Expand All Advertisement Advertisement PDF downloadLoading ...

Ähnliche Arbeiten